Performance of a convolutional autoencoder designed to remove electronic noise from p-type point contact germanium detector signals

TL;DR

This paper introduces a convolutional autoencoder that effectively denoises signals from germanium detectors, improving energy resolution and signal shape preservation, with applications in particle physics and other fields.

Contribution

The study presents a novel autoencoder-based denoising method that works with both simulated and real detector data, enhancing signal quality and analysis capabilities.

Findings

Denoising preserves pulse shape and improves energy resolution.

Method reduces required shaping time for energy calculation.

Autoencoder approach is adaptable to other detector technologies.

Abstract

We present a convolutional autoencoder to denoise pulses from a p-type point contact high-purity germanium detector similar to those used in several rare event searches. While we focus on training procedures that rely on detailed detector physics simulations, we also present implementations requiring only noisy detector pulses to train the model. We validate our autoencoder on both simulated data and calibration data from an $^{241}$Am source, the latter of which is used to show that the denoised pulses are statistically compatible with data pulses. We demonstrate that our denoising method is able to preserve the underlying shapes of the pulses well, offering improvement over traditional denoising methods. We also show that the shaping time used to calculate energy with a trapezoidal filter can be significantly reduced while maintaining a comparable energy resolution. Under certain circumstances, our denoising method can improve the overall energy resolution. The methods we developed to remove electronic noise are straightforward to extend to other detector technologies. Furthermore, the latent representation from the encoder is also of use in quantifying shape-based characteristics of the signals. Our work has great potential to be used in particle physics experiments and beyond.